THERA-IE: An AI-Enabled System for Therapeutic Indication Identification and Extraction from Biomedical Literature.

Area of Science:

• Computational Biology
• Bioinformatics
• Natural Language Processing

Background:

• Maintaining comprehensive drug-indication knowledge is crucial for evidence-based prescribing but is challenging due to incomplete data, especially for off-label uses.
• Biomedical literature and artificial intelligence (AI) models hold potential for capturing this underrepresented therapeutic information.

Purpose of the Study:

• To present THERA-IE (Therapeutic Hypothesis Extraction and Relationship Analytics - Indication Extraction), an open-source framework for identifying putative therapeutic indications.
• To evaluate the performance of large language models (LLMs) and other AI approaches in extracting drug-indication relationships from text.

Main Methods:

• Utilized SemMedDB, PubMedBERT, and three open-source LLMs (Llama 3.2, MedLlama 3, Qwen 3) to analyze twenty FDA-approved drugs.
• Implemented LLMs in both model-only ('Naïve') and literature-supported ('Literature-Based') modes.
• Benchmarked system performance against DrugBank, measuring metrics like F1-score, precision, and recall.

Main Results:

• Naïve Llama 3.2 achieved the highest overall F1-score (0.60 ± 0.18).
• Literature-Based Llama 3.2 demonstrated the highest precision (0.67 ± 0.29).
• SemMedDB yielded the highest recall (0.77 ± 0.25) but also the most false positives. THERA-IE identified an average of 4.8 plausible off-label indications per drug.

Conclusions:

• The THERA-IE framework effectively extracts therapeutic knowledge, including potential off-label indications, from biomedical literature.
• AI, particularly LLMs, offers a scalable and reproducible approach to supplement existing drug-indication databases.
• This work highlights the potential for automated knowledge extraction to advance evidence-based medicine.